Mapping maintenance for data management

Despite the potential of data integration systems, their deployment in practice is still quite limited. Building such a system is an expensive endeavor, for which numerous semiautomatic tools have been developed. Once the system has been built and deployed, an equally daunting challenge is to maintain it over time. In dynamic environments (such as the Web), sources often autonomously change without regard for the system, resulting in the need to continually monitor, detect, and repair broken components. One particularly susceptible component is the set of semantic mappings between the global schema and the schemas of data sources. In this thesis, MAVERIC, an automatic mapping verification system is presented. MAVERIC periodically probes a data integration system, and alerts the system administrator if a mapping has become broken. The core architecture of MAVERIC: a collection of inexpensive sensors that are trained and deployed to verify the mappings is described. Then three novel improvements are developed - perturbation and multi-source training to make the verification system more robust, and filtering to reduce the number of false alarms. We present extensive experiments over 111 real-world sources in six domains. The results demonstrate the effectiveness of our core approach over existing solutions, as well as the utility of our improvements.

Name thesis:degree_name

M.S. (master's)

Level thesis:degree_level

Thesis

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2025

Author dc:creator

• Al-Shebli, Bedoor Kh.

Contributors dc:contributor

• Kelley, Mary Beth

• Computer science
• Data management

Language dc:language

eng